Summaries of public announcements, with links to the original source. Updated monthly.
Am I Ready for Agents? editors · Reviewed September 2026
Most 2026 news about AI agents and websites came from infrastructure companies and standards projects. Cloudflare shipped new controls for AI crawlers and agents, the main agent standards settled into shared foundations, and ora.ai released scanning tools. Each item says what happened and why a business reader might care.
MCP server portals give a company one approved entry point for the MCP servers its staff and agents use, with logging of tool activity. For a business, it is a sign that controlling which agent tools are allowed is becoming a standard IT task.
Source: Cloudflare changelog
A new Disallow AI Training setting writes robots.txt rules that refuse training while keeping a site discoverable in search. Cloudflare also introduced an Accountable label for crawler operators that meet stated opt-out and reporting requirements.
Source: Cloudflare blog
Operators of bots and agents now submit them through a BotBase dashboard, and Cloudflare validates verification methods such as Web Bot Auth signatures automatically. It should make it easier for sites to tell a real agent from one using a borrowed name.
Source: Cloudflare blog
A2A, which lets agents from different organisations work together, joined the Agentic AI Foundation, which it describes as directed by the Linux Foundation and home to MCP too. Standards with a shared home are a safer long-term bet for businesses.
Source: A2A project blog
Cloudflare introduced optional OAuth scopes, so a person can untick permissions they do not want to grant on the consent screen, citing agent tools that request broad access. For booking and account-based businesses, it points to consent that matches the task.
Source: Cloudflare blog
is-agentic.com, built by Vercel with scans run by ora.ai, takes a URL and returns a score with the fixes to raise it, split into essential, recommended and bonus checks.
Source: ora.ai blog
Version 2 lets any page point to its markdown version and to the llms.txt file that covers it, so an agent arriving on a deep page can find the site's guide. A version 1 file remains a valid starting point.
Source: llmstxt.org
An Agents Week post describes agents as a new kind of visitor and groups the infrastructure they need into four ideas: readable, discoverable, callable and payable. It is a useful map of where CDN defaults are heading.
Source: Cloudflare blog
A developer preview adds WebMCP tools to pages at the edge, so browser agents can call them without changes to the site itself. It is early, but it shows agent features arriving through infrastructure rather than rebuilds.
Source: Cloudflare blog
The 2026-07-28 specification removes the opening handshake and session IDs, making connections stateless, and tightens how authorization works. For businesses offering an MCP server, it is the version to build on.
Source: MCP blog
ora.ai says Deep Scan v2 builds its checklist from agents run on real tasks, and separates Verified signals, backed by agent behaviour, from Emerging signals that are tracked but not yet scored.
Source: ora.ai blog
The x402 project introduced batch settlement, which it says lets agents make very small payments quickly and cheaply using vouchers redeemed in bulk. Agent payments are moving, but no approach is yet universal.
Source: x402.org
ora.ai described Deep Scan as a benchmark rather than a checklist: instead of only confirming that files exist, it runs agent tasks, live sign-in flows and integration tests.
Source: ora.ai blog
Markdown for Agents converts a page to markdown at the edge when an agent asks for it, at no cost on Pro, Business and Enterprise plans. Cloudflare's documentation adds that sites without their own policy send a content signal saying yes to AI training, search and AI input.
Source: Cloudflare blog · Cloudflare docs